About

Analysis of Variance (ANOVA) determines whether the means of 3 or more independent groups differ significantly. A miscalculated F-statistic leads to Type I errors (false positives) or Type II errors (missed real effects). Both cost time and credibility in peer review or production decisions. This tool computes the full one-way ANOVA table: SS_B, SS_W, SS_T, degrees of freedom, Mean Squares, the F-ratio, and an exact p-value via the regularized incomplete beta function. It assumes independent observations, approximate normality within groups, and homogeneity of variances (Levene's assumption).

Beyond the omnibus test, a significant F only tells you "at least one group differs." You need post-hoc analysis to find which pairs. This calculator runs Tukey HSD pairwise comparisons and reports effect size via η² (eta-squared) and ω² (omega-squared, bias-corrected). Note: the tool approximates the Studentized Range critical value q using interpolation. For unbalanced designs (unequal group sizes), the harmonic mean of group sizes is used in the HSD formula. Results assume α = 0.05 by default but the threshold is adjustable.

Formulas

The one-way ANOVA partitions total variability into two components. The F-statistic tests the null hypothesis H₀: μ₁ = μ₂ = … = μ_k.

F = MS_BMS_W = SS_B ÷ (k − 1)SS_W ÷ (N − k)

Where SS_B = k∑j=1 n_j(X_j − X_G)² and SS_W = k∑j=1 n_j∑i=1 (x_ij − X_j)².

k = number of groups. n_j = number of observations in group j. N = total number of observations across all groups. X_j = mean of group j. X_G = grand mean of all observations. p = probability of the observed F under the null hypothesis, computed via the regularized incomplete beta function I_x(a, b).

The Tukey HSD critical difference for unequal group sizes uses the harmonic mean n_h: HSD = q_{α,k,df_W} ⋅ √MS_Wn_h. A pairwise absolute mean difference |X_i − X_j| exceeding HSD indicates a statistically significant pair.

Reference Data

Source	Symbol	Formula	Interpretation
Grand Mean	X̄_G	N∑i=1 x_iN	Overall mean of all observations
SS Between	SS_B	k∑j=1 n_j(X̄_j − X̄_G)²	Variation due to group differences
SS Within	SS_W	k∑j=1 n_j∑i=1 (x_ij − X̄_j)²	Variation within groups (error)
SS Total	SS_T	SS_B + SS_W	Total variation in the dataset
df Between	df_B	k − 1	Number of groups minus one
df Within	df_W	N − k	Total observations minus groups
MS Between	MS_B	SS_Bdf_B	Average variation between groups
MS Within	MS_W	SS_Wdf_W	Average variation within groups
F-Statistic	F	MS_BMS_W	Ratio of between to within variance
P-Value	p	1 − F_cdf(F, df_B, df_W)	Probability of observing F under H₀
Eta-Squared	η²	SS_BSS_T	Proportion of total variance explained
Omega-Squared	ω²	SS_B − df_B ⋅ MS_WSS_T + MS_W	Bias-corrected effect size estimate
Tukey HSD	HSD	q_{α,k,df_W} ⋅ √MS_Wn_h	Minimum difference for significance
Effect Size Guide	η²	0.01 small, 0.06 medium, 0.14 large	Cohen's benchmarks for ANOVA
Harmonic Mean	n_h	kk∑j=1 1n_j	Used in HSD for unequal group sizes

Frequently Asked Questions

Three core assumptions apply. First, independence: observations within and between groups must be independent of each other. Second, normality: the data within each group should be approximately normally distributed, though ANOVA is robust to moderate violations when group sizes are ≥ 20. Third, homogeneity of variances (homoscedasticity): the variances across groups should be roughly equal. Levene's test can verify this. If variances are heterogeneous, consider Welch's ANOVA instead.

The omnibus F-test only evaluates whether at least one group mean differs from the others. It aggregates all between-group variation into a single ratio. To identify which specific pairs of groups are significantly different, you need a post-hoc test such as Tukey HSD, which controls the family-wise error rate across all pairwise comparisons. This calculator performs Tukey HSD automatically when the F-test is significant.

Unbalanced designs (unequal n_j) do not invalidate ANOVA, but they reduce statistical power and make the test more sensitive to violations of homogeneity of variances. For Tukey HSD in unbalanced designs, this calculator uses the Tukey-Kramer modification, which substitutes the harmonic mean of the two group sizes being compared. This approach maintains the correct Type I error rate.

Eta-squared (η² = SS_B / SS_T) is the proportion of total variance explained by group membership. It is positively biased, especially with small samples. Omega-squared (ω²) corrects this bias by subtracting df_B × MS_W from SS_B in the numerator and adding MS_W to SS_T in the denominator. For publication-quality reporting, omega-squared is preferred.

Yes, but it is mathematically equivalent to an independent-samples t-test. The F-statistic will equal t², and the p-values will be identical. ANOVA becomes useful with 3 or more groups because it avoids the inflated Type I error rate that results from running multiple pairwise t-tests (e.g., 3 groups require 3 comparisons, each at α = 0.05, yielding a combined error rate of approximately 0.14).

The p-value is derived from the cumulative distribution function (CDF) of the F-distribution. This calculator evaluates the CDF using the regularized incomplete beta function I_x(a, b), implemented via a continued fraction expansion (Lentz's algorithm). The relationship is: p = 1 − I_x(df_B/2, df_W/2) where x = df_B · F / (df_B · F + df_W). This is accurate to approximately 10 decimal places for typical inputs.